ISSN 1000-1239 CN 11-1777/TP

计算机研究与发展 ›› 2018, Vol. 55 ›› Issue (7): 1508-1524.doi: 10.7544/issn1000-1239.2018.20170252

• 信息处理 • 上一篇    下一篇



  1. 1(中国科学院大学 北京 101408); 2(中国科学院软件研究所 北京 100190) (
  • 出版日期: 2018-07-01
  • 基金资助: 

A Feature-Based Co-Clustering Model

Zhang Fei1, Zhang Libo2, Luo Tiejian1, Wu Yanjun2   

  1. 1(University of Chinese Academy of Sciences, Beijing 101408); 2(Institute of Software, Chinese Academy of Sciences, Beijing 100190)
  • Online: 2018-07-01

摘要: 推荐系统能够有效解决用户的个性化推荐问题,其中,协同过滤是近年来的主流方法.协同过滤算法具有一定的局限性,因为需要在全部的物品中为用户进行推荐,而单个用户往往只对某些领域的物品感兴趣.为了解决这个问题,提出了一种新的协同聚类模型,先将用户和物品根据兴趣或特征进行聚类分组,然后在每个分组上进行相应的推荐.该模型主要包含2个模块:1)特征表示模块,用以发掘用户的兴趣和物品的特征;2)根据该特征构建的图模型,用来求解最终的聚类分组.通过在3种公开数据集上与其他算法进行性能比较,验证了这种协同聚类模型能够显著提高推荐系统预测与推荐的准确度.

关键词: 推荐系统, 聚类模型, 协同过滤, 特征表示, 图模型

Abstract: Recommendation system can effectively solve the personalized recommendation problem for users. As one of the most commonly used algorithm in recommendation system, collaborative filtering needs to take all the items into account, while a specific user may be only interested in the items in some certain domains. It’s more natural to make recommendation for a user via the correlated domains than the entire items, therefore, users and items can be grouped according to their interests or characteristics, and then the recommendations can be made with the user-item subgroups. Based on this idea, we propose a novel co-clustering method based on the features of users and items to find the meaningful subgroups. The proposed method includes two main modules: a feature representation model to explore the interests of the users and the characteristics of the items, and a graph model constructed in accordance with these features for coming up with the final clustering results which are used for making recommendation. In this paper, we introduce the framework of our method and give an effective solution to get the features and the clustering results. Finally, by comparing with a variety of newest algorithms on three open datasets, we verify that the proposed method can significantly improve the accuracy of recommender system.

Key words: recommender system, clustering model, collaborative filtering, feature representation, graph model